1. Field of the Invention
The present invention relates to a data compression system and to concealing errors in compressed data. An embodiment of the invention concerns concealing errors in compressed data reproduced from a recording apparatus. In a preferred embodiment the data is transformed by a two dimensional spatial frequency transform such as a wavelet transform. The background to the present invention, and the present invention are described herein, by way of example, with reference to wavelet transformed video data.
2. Description of the Prior Art
The wavelet transform is known. Referring to FIG. 1, data subjected to the transform occupies a plurality of two-dimensional spatial frequency bands. In the example of FIG. 1 there are seven bands denoted 0 to 6. There may be other numbers of bands e.g. 9 or more. The wavelet transform transforms the data such that the video data in bands 4, 5 and 6 is sub-sampled horizontally and vertically by 1/2 and so represent 1/4 size images. The video data bands 0, 1, 2 and 3 is sub-sampled horizontally and vertically by 1/4 and so represents 1/16 size images. Horizontal spatial frequency increases generally from left to right as indicated by arrow H and vertical spatial frequency increases generally from top to bottom as indicated by arrow V. It is found that typically the most significant image data is in the lowest frequency band 0 and the least significant image data is in the highest frequency band 6.
Video data, as very well known, may be recorded on a medium such as tape. The recording may introduce errors in the data, due to a variety of factors including tape defects and head failure for example. It is known to detect the errors and to conceal them. Concealment is generally done by replacing data in error by some form of average of adjacent data: that is a form of low pass filtering.
The wavelet transform has the property that data in bands 1 to 6 is frequency inverted after transformation. Low pass filtering does not always produce the best replacement value for data errors when applied to wavelet transformed data having frequency inverted data.